结合区域全局特征和Ann-SIFT的二阶段快速景象匹配算法  

Two-stage fast scene matching algorithm combining global descriptor and Ann SIFT

作　　者：辛瑞 张霄力 彭侠夫 陈锦文 Xin Rui;Zhang Xiaoli;Peng Xiafu;Chen Jinwen(School of Aerospace Engineering,Xiamen University,Xiamen 361005,China)

机构地区：[1]厦门大学航空航天学院,福建厦门361005

出　　处：《电子技术应用》2023年第5期135-141,共7页Application of Electronic Technique

基　　金：航空科学基金(201958068002)。

摘　　要：在GPS拒止环境下,提出了一种基于高清航拍图像的视觉定位方法。为了保证高分辨率图像匹配的实时性,结合粗粒度高级语义全局特征和细粒度点特征实现两阶段快速景象匹配。第一阶段,设计三元组网络针对图像区域提取具有尺度和旋转不变性的全局描述符进行区域预匹配。第二阶段,快速最近邻点搜索的Ann-SIFT对实时图和参考区域进行精确点匹配以实现定位。对于3 000×3 000的参考图像和500×500到1 000×1 000尺寸范围的实时图像,平均耗时可以从66.24 s降低到0.97 s。Under the circumstance of GPS-denied environment,a visual positioning method based on high-definition aerial images is proposed.In order to ensure the real-time performance of high-resolution image matching,two-stage fast scene matching is realized by combining coarse-grained high-level semantic global features and fine-grained point features.In the first stage,a triplet network is designed to extract scale and rotation-invariant global descriptors for referance image blocks for region pre-matching.In the second stage,Ann-SIFT for fast nearest neighbor search performs precise point matching on the real-time image and reference area to achieve localization.For reference images of 3000×3000 and real-time images in the size range from 500×500 to 1000×1000,the average time-consuming can be reduced from 66.24 s to 0.97 s.

关 键 词：景象匹配 视觉定位 深度学习 全局描述符 

分 类 号：TN967.2[电子电信—信号与信息处理]